We model the history of Galactic r -process enrichment using high-redshift , high-resolution zoom cosmological simulations of a Milky Way ( MW ) type halo .
We assume that all r -process sources are neutron star mergers ( NSMs ) with a power law delay time distribution .
We model the time to mix pollutants at subgrid scales , which allows us to to better compute the properties of metal poor ( MP ) and carbon enhanced metal poor ( CEMP ) stars , along with statistics of their r -process enhanced subclasses .
Our simulations underpredict the cumulative ratios of r -process enhanced MP and CEMP stars ( MP- r , CEMP- r ) over MP and CEMP stars by about one order of magnitude , even when the minimum coalescence time of the double neutron stars ( t _ { min } ) is set to 1 Myr .
No r -process enhanced stars form if t _ { min } = 100 Myr .
Our results show that even when we adopt the r -process yield estimates observed in GW170817 , NSMs by themselves can only explain the observed frequency of r -process enhanced stars if the birth rate of double neutron stars per unit mass of stars is boosted to \approx 10 ^ { -4 } M _ { \odot } ^ { -1 } .